Build security environments, adversarial agents, and evaluations for frontier AI systems.
Location: Tübingen / Hybrid
Employment: Internship
The role
You will work across AI security, machine learning, and research engineering alongside the core team. You will take ownership of a technically ambitious project in agent security, from designing and implementing the approach to running experiments and communicating the results.
What you’ll work on
Develop and evaluate security environments, agent training methods, attacks, or defenses
Run controlled experiments with clear baselines and verifiable outcomes
Analyze model behavior and communicate the results to the team
Contribute code, evaluations, and research findings that remain useful after the internship
What we’re looking for
Current study or recent work in machine learning, computer science, security, or a related field
Strong programming skills and experience building or evaluating machine-learning systems
Curiosity about AI agents and the discipline to investigate failures carefully
Ability to work independently and ask for feedback early
Useful experience
LLMs, reinforcement learning, or agentic systems
Adversarial machine learning, red teaming, or AI safety evaluations
Research codebases, reproducible experiments, or open-source contributions
A previous research project, thesis, or technical publication
What we offer
Competitive compensation
Flexible hours and a remote-friendly setup
Vacation, learning, and conference budget
Research freedom in a high-trust environment
Work with us
Submit your CV, a short note, and links to work that shows how you think and build using the application form.